Application of Improved YOLOv5 in Infrared Image Recognition of Electrical Equipment

Infrared detection is one of the methods commonly used to detect thermal faults in power equipment. Fast and accurate recognition of equipment in infrared images is the key to avoiding thermal faults. At present, the use of a large number of infrared supervisory equipment has made the number of collected infrared images grow exponentially. In these infrared images, the identification of electric power equipment is still basically relying on manual work, which has the disadvantages of high work intensity and low efficiency. In this paper, we improve the YOLOv5(You Only Look Once) algorithm for accomplishing the identification and localization of equipment types in infrared images. Before detection, the input infrared images are preprocessed to reduce the influence of noise, which effectively improves the detection accuracy. In addition, an ASPPF module is introduced to address the information loss caused by maximum pooling. The proposed method includes the tuning of the model parameters, the optimization of the training strategy, and then the optimal weights are obtained; subsequently, a huge number of infrared images collected from the converter station are used to train and validate the model. Through experimental validation, it is found that our improved algorithm has a significant improvement in recognition accuracy compared with the commonly used detection algorithms.